PaperlayerView exact revision on arXiv

Source-linked AI summary

Syntax Element Encryption for H.265/HEVC Using Chaotic Map-Based Coefficient Scrambling Scheme

Liang-Wei Li, Chung-Nan Lee, Kishu Gupta, Huei-Fang Yang, Ashutosh Kumar Singh

arXiv:2608.22573v1cs.CR

TL;DR

Existing H.265/HEVC coefficient-scrambling approaches leave bitrate overhead and efficiency concerns while requiring secure selective encryption. The paper proposes chaotic-map coefficient scrambling combined with encryption of MVD, QTC, and Luma IPM, using slice synchronization and AES-CTR-based randomness. The scheme reports low encryption-time overhead, reduced average bitrate overhead relative to RSVE, and effective visual distortion with reversible decryption, although encrypted-frame entropy remains lower than desired.

  • Problem

    Existing coefficient-scrambling methods present bitrate-overhead, execution-time, and security concerns, motivating a more efficient H.265/HEVC encryption approach.

  • Method

    The paper combines chaotic-map coefficient scrambling with selective encryption of MVD, QTC, and Luma IPM, using AES-CTR randomness synchronized to slices.

  • Results

    1.91% average encryption-time overhead versus approximately 17.05% for RSVE, while decrypted frames are identical to originals and encrypted frames are difficult to visually distinguish from them.

  • Takeaways & Limitations

    The proposed scheme offers format-compliant, slice-synchronized H.265/HEVC encryption with fast execution, reduced average bitrate overhead relative to RSVE, and resistance to common attacks.

  • Takeaways & Limitations

    Encrypted-frame entropy is lower, suggesting incomplete ciphertext randomness because the chaotic map may introduce statistical bias and residual patterns.

Abstract

from arXiv · show

In today's digital landscape, high-efficiency video coding (H.265/HEVC) has emerged as the most widely used video coding standard, employing selective encryption schemes to protect the privacy of video content while maintaining efficient compression performance. However, existing coefficient scrambling methods impose a significant computational load, leading to increased bit rate overhead due to encryption, longer execution times, and insufficient safety measures. To address these issues, a new coefficient scrambling scheme based on \textit{chaotic maps} is proposed. This approach leverages the pseudorandomness, ergodicity, and sensitivity to initial conditions inherent in chaotic maps to generate highly unpredictable coefficient distributions, thereby strengthening security while preserving low complexity. Unlike conventional scrambling, chaotic maps ensure minimal correlation between encrypted coefficients, enhancing resistance against statistical and differential attacks. Additionally, the scrambling conditions are specifically designed to minimize the impact on the bit rate overhead. Furthermore, when combined with syntax element encryption (SEC), which includes motion vector difference (MVD), quantized transform coefficients (QTC), and luma intraprediction mode (Luma IPM), this method effectively distorts video content. The proposed scheme operates synchronously with slices, ensuring that the decryption of video content remains intact even if some slices are lost. Additionally, a random sequence generated by AES-CTR is incorporated with the H.265 encoded stream to protect against chosen-plaintext attacks.

I. INTRODUCTION

The paper proposes an H.265/HEVC selective-encryption scheme combining chaotic-map coefficient scrambling with encryption of selected syntax elements. It targets video security while reducing scrambling-related bitrate overhead and preserving slice-synchronized decryption.

  • Security and synchronization: AES-CTR generates the random sequence, which is synchronized with slices and linked to the H.265 encoded stream.The slice synchronization is intended to preserve decryption when some slices are lost.
  • Proposed scheme: The scheme encrypts MVD, QTC, and Luma IPM information while maintaining H.265/HEVC format compliance.The selected syntax elements are intended to distort video content without preventing decoder operation.
  • Proposed scheme: Chaotic maps generate coefficient scrambling sequences that disrupt coefficient distributions through pseudorandomness.The approach is presented as an efficient alternative for disturbing encrypted coefficient structure.
  • Design objectives: Scrambling conditions are designed to protect video edges and minimize coefficient-scrambling bitrate overhead.This design addresses the overhead problem identified for existing scrambling methods.
  • Reported evaluation: The paper reports theoretical and experimental improvements across performance metrics compared with state-of-the-art methods.The stated contributions include reductions in bitrate and execution-time overhead and resistance to substitution attacks.
  • CABAC background: CABAC converts input syntax elements into bin strings, applies context modeling or bypass coding, and then performs binary arithmetic coding.Binarization may be omitted when a syntax element is already represented in binary.

C. Chaotic Maps for Pseudorandom Values

The proposed scheme generates pseudorandom values with a low-dimensional logistic-sine chaotic system and uses them to drive slice-synchronous coefficient scrambling alongside selective syntax-element encryption.

  • Workflow: The scheme combines chaotic-map coefficient scrambling with selective encryption of syntax elements within the H.265 encoding and CABAC workflow.Selected syntax elements are converted to binary strings before encryption, and the encrypted stream is produced after encoding.
  • Slice-synchronous randomness: Each slice updates its key and IV from a slice-header hash, then AES-CTR generates random sequences for coefficient scrambling and syntax-element encryption.The slice key uses the first 256 hash bits, while the slice IV uses the last 128 bits; AES-CTR produces 64-bit random sequences.
  • Chaotic pseudorandom generation: The logistic-sine low-dimensional chaotic system combines logistic and sine dynamics to enhance ergodicity, initial-condition sensitivity, and statistical randomness without higher-dimensional computational burden.The system is selected specifically to avoid excessive execution time.
  • Chaotic pseudorandom generation: With µ = 7.3 and x(0) = 0.9324, the system iterates 65,535 times to produce 65,536 pseudorandom values.The parameter range is 0 ≤ µ ≤ 9, and x(n) generates values in R.
  • Coefficient scrambling: Coefficient scrambling selects a coefficient group, removes specified patterns, reads pseudorandom values, sorts them into indices, and shifts coefficients according to those indices.The number of scrambles is S = mod(S_i, N − 1) + 2, after which indices are extracted and used for right shifts.
  • Coefficient scrambling: The proposed perturbation is described as producing a more random coefficient distribution than exchange-based or two-round-shift alternatives.The cited comparison attributes limitations to random-object selection and restricted perturbation distributions in earlier schemes.

1) MVD Information:

MVD encryption separately processes sign information and suffix information, while QTC coding uses a rice parameter whose changes can threaten format compliance.

  • MVD sign and suffix encryption: MVD horizontal and vertical sign flags are encrypted through separate operations.The paper identifies these operations as MVD Hor Sign and MVD Ver Sign.
  • MVD sign and suffix encryption: MVD suffixes are encrypted after representing the remaining level as a zero-prefix, one-bit, and N-bit suffix structure.Horizontal and vertical remaining MVD suffixes use separate encryption equations.
  • QTC information: QTC coefficient signs are encrypted separately from the remaining coefficient suffix.The QTC remaining-level structure uses an M−1 ones prefix, a zero, and an M-bit suffix.
  • QTC information: The rice parameter r is updated to min(4, r + 1) when QTC size exceeds 3 × 2^r.This condition is given as the rule for computing r.
  • QTC information: Encrypting the remaining coefficient suffix requires conditions on encryption eligibility and the number of encryption bits L because it may alter the rice parameter and cause format-compliant failure.The suffix is encrypted through the specified equation only when those conditions permit encryption.

3) Luma IPM Information:

The section integrates luma-IPM encryption with syntax-element scrambling while preserving decoding relationships among prediction modes and coefficients. It also presents theoretical security and chaotic-dynamics arguments for the proposed scheme.

  • Luma IPM Information:: Luma IPM contains 35 prediction modes, including planar, DC, and 33 angle modes, and encryption is performed through Eq. (12).
  • Luma IPM Information:: Direct luma-IPM encryption can cause format-compliance failure, so the original and encrypted luma IPMs are recorded before additional processing.
  • Luma IPM Information:: Encrypted luma IPM determines adjacent-PU candidate modes, coefficient scan mode, and chroma-IPM associations needed for decoding.
  • Luma IPM Information:: Chroma-IPM adjustment is decided before encoding because encrypted luma IPM can otherwise change the decoded chroma mode and scan-mode decisions.
  • Luma IPM Information:: The security proof models chaotic scrambling as a random permutation and concludes negligible CPA advantage under CMPA and PRF assumptions.The proof uses a simulator that encrypts one of two chosen coefficient sets and distinguishes real chaotic permutations from random permutations.
  • Luma IPM Information:: The Logistic-Sine Chaotic Map provides parameter sensitivity, a validated chaotic region, and a positive Lyapunov exponent for unpredictable pseudorandom sequences.For µ ∈[6, 9], numerical evaluation yields λ > 0; small changes in initial conditions or parameters produce exponentially divergent sequences.

C. Theoretical Proof for Information Entropy Analysis

The entropy analysis argues that chaotic coefficient scrambling increases randomness by transforming the concentrated residual-coefficient distribution toward uniformity. This reduces statistical structure in encrypted coefficients, depending on the scrambling operation.

  • Theoretical Proof for Information Entropy Analysis: The proposed coefficient scrambling scheme is intended to increase syntax-element entropy and resistance to statistical and entropy-based attacks.
  • Theoretical Proof for Information Entropy Analysis: Original HEVC residual coefficients follow a Laplacian-like distribution concentrated near zero, which produces lower entropy.
  • Theoretical Proof for Information Entropy Analysis: 5.5 bits/symbol is the empirical entropy H(C) for 8-bit HEVC coefficients, compared with the 8-bit maximum entropy.
  • Theoretical Proof for Information Entropy Analysis: Permutation-only scrambling preserves the histogram and entropy while destroying spatial correlation, whereas chaotic value masking produces an approximately uniform distribution.

D. Security Proof

The security proof characterizes the scheme through key-space size, entropy, confusion and diffusion, mutual-information reduction, and CPA resistance. These claims rely on chaotic-map and SHA-384 pseudorandomness assumptions.

  • Security Proof: The scheme is presented as providing information-theoretic security through a large key space, high entropy, confusion, diffusion, and CPA resistance.
  • Security Proof: The total key space exceeds 2^300, based on x0 ∈(0, 1), µ ∈[6, 9] at 10^-15 precision, and SHA-384 key updates.
  • Security Proof: LSCM supplies nonlinear confusion, while dynamic SHA-384 key updates propagate changes across syntax elements to provide diffusion and an avalanche effect.
  • Security Proof: The CPA-resistance claim assumes LSCM outputs are computationally indistinguishable from random and SHA-384 behaves as a pseudorandom function.
  • Security Proof: The mutual-information analysis states that ciphertext reveals negligible information about plaintext.

11 End

The section defines image-quality, randomness, differential-security, edge-difference, and bitrate-overhead metrics for evaluating encrypted video. These metrics quantify distortion, sequence differences, entropy, and coding-cost impact.

  • Performance Metrics: PSNR and SSIM are used to evaluate similarity between original and encrypted frames, with lower values indicating lower similarity.PSNR is measured in dB, while SSIM ranges from 0 to 1.
  • Performance Metrics: EDR measures edge-frame differences from 0 to 1, with higher values indicating greater differences between edge frames.
  • Performance Metrics: Bitrate overhead measures encryption-induced file-size increase from the original video, with 0% representing constant bitrate.The metric uses encrypted-video file size ES and original-video file size OS.
  • Performance Metrics: HDavg measures differences between two random sequences, and values closer to 0.5 indicate greater differences relevant to CPA resistance.
  • Performance Metrics: Information entropy measures randomness; for an 8-bit grayscale image, the maximum is 8, and values closer to 8 indicate greater randomness.
  • Performance Metrics: NPCR and UACI evaluate differential changes between cipher frames using pixel-value differences at corresponding positions.The expected NPCR and UACI values are computed from the cipher-frame pixel comparisons.

VI. PERFORMANCE EVALUATION

The evaluation uses HM17.0 random-access main encoding across 12 video sequences with resolutions from 352×288 to 2560×1600, under fixed experimental settings.

  • Experimental Setup: The experiments use HM17.0 random-access main encoding with QP 24, 100 encoded frames, and SideHideFlag disabled.The experimental environment and configuration are reported in Table II.
  • Video Sequences: The test set contains 12 video sequences spanning resolutions from 352×288 to 2560×1600.Results are compared with the robust RSVE scheme.
  • Comparison: The proposed scheme is evaluated against RSVE across the selected video sequences.The comparison is conducted under the stated HM17.0 encoding configuration.

B. Experimental Results

The proposed scheme strongly distorts encrypted video while preserving exact decryption, supports slice-synchronized recovery, and resists chosen-plaintext and substitution attacks. It also achieves strong objective distortion and edge hiding, although RSVE performs better on visual perception and entropy-related measures reported elsewhere.

  • Subjective Vision Analysis: The decrypted frames are identical to the originals, while encrypted frames make texture and outlines difficult to distinguish.This behavior is reported for the Johnny and BasketballDrive sequences.
  • Slice Analysis: After losing slices 11–14, the decrypted 15th frame remains identical to the corresponding original 18th frame.The result is supported by matching PSNR and SSIM values and indicates slice-synchronized random-sequence generation.
  • Subjective Vision Analysis: The proposed method produces distortion comparable to RSVE while maintaining correct decryption.The subjective-analysis comparison uses original, RSVE, proposed, and decrypted frames.
  • Objective Indicator Analysis: The proposed scheme yields significantly lower PSNR and second-best SSIM among the compared methods.The results are averaged over 100 frames; RSVE provides better visual perception despite the proposed scheme’s stronger image distortion.
  • Edge Detection Analysis: Encrypted edge frames are strongly distorted, with EDR values close to 1 and slightly better performance than RSVE.The reported advantage is attributed to the coefficient-scrambling conditions.
  • Chosen-Plaintext Attack Analysis: The proposed scheme resists chosen-plaintext attacks because slice-header updates produce different random sequences for each slice.HDavg values are reported as close to 0.5 for video groups with the same resolution.
  • Replacement Attack Analysis: Four substitution attacks produce decrypted frames with smaller PSNR and SSIM values than the original frames, supporting attack resistance.The tested substitutions include encrypted Luma IPM and MVD suffix replacements, with results reported for PartyScene.

I. Information Entropy Analysis

The analysis evaluates information entropy, bit-rate overhead, brute-force resistance, and encryption-time overhead. The proposed scheme maintains substantial pixel-distribution randomness and improves overhead and execution time relative to RSVE, while RSVE has higher entropy.

  • Information Entropy Analysis: The proposed scheme maintains a significant level of randomness in pixel distributions, although RSVE achieves superior information-entropy indices.Entropy results are reported over 100 frames for various videos.
  • Bit-Rate Overhead Analysis: 45.13% lower average bit-rate overhead is reported for the proposed scheme compared with RSVE when coefficient scrambling is included.The comparison is presented in Table X.
  • Brute-Force Attack Analysis: A 256-bit AES key provides 2^256 possible combinations for brute-force attacks.The paper uses AES to generate the random sequence and describes this key space as adequate against brute-force attacks.
  • Encryption Time Overhead Analysis: The average encryption-time overhead is about 1.91% for the proposed scheme versus approximately 17.05% for RSVE.The proposed scheme performs comparably to RSVE across 12 videos and is reported as efficient for low-resolution sequences.
  • Encryption Time Overhead Analysis: The proposed algorithm demonstrates improved performance for low-resolution video sequences and is described as suitable for mobile devices.This conclusion accompanies the encryption-time overhead analysis.

M. NPCR and UACI Analysis

The proposed scheme is compared across NPCR, UACI, computational complexity, and broader security-practicality criteria. It achieves optimal NPCR and commendable UACI, while the discussion acknowledges mixed performance across metrics and identifies targeted improvements as future work.

  • NPCR and UACI Analysis: The proposed scheme outperforms the compared approaches in NPCR and has UACI results close to the expected value.Tables XII and XIII report results across various video sequences and state-of-the-art methods.
  • Computational Complexity: The computational-complexity comparison orders the proposed scheme alongside selected state-of-the-art methods.Table XIV defines complexity parameters including processed syntax elements, extracted features, hashing-related parameters, and encryption hierarchy.
  • Comparative Analysis: The comparative evaluation uses six criteria covering format compliance, encrypted-video distortion, execution-time overhead, and bit-rate overhead.Lower overhead is preferred for real-time or large-scale applications, while minimal bit-rate increase preserves storage and transmission efficiency.
  • Discussion and Limitations: The proposed scheme does not consistently outperform existing approaches across all evaluated metrics.The discussion identifies mixed performance and trade-offs among quality, security, and efficiency measures.
  • Discussion and Limitations: Future work targets SSIM, information entropy, and UACI through hybrid diffusion, adaptive coefficient selection, and additional chaotic transformations.These refinements are intended to improve consistency while retaining current advantages in PSNR, edge preservation, NPCR, and bit-rate efficiency.
  • Conclusion and Future Work: The conclusion reports high security, format compliance, fast execution, slice synchronization, attack resistance, and reduced average bit-rate overhead relative to RSVE.The paper identifies future work involving a 2D extended Schaffer function map, neural networks, and temporal action segmentation.
Loading 2608.22573v1…